Recall from the sentence immediately preceding $(10.20)$ that $\rho$ is a state in the code subspace $C$ and from the statement of theorem $10.1$ that $P$ is a projector onto $C$. We will show that $\sqrt{\rho}=P\sqrt{\rho}$.
Let $|i_L\rangle$ be an orthonormal basis of the code subspace. Then
$$
P=\sum_i|i_L\rangle\langle i_L|.\tag1
$$
Therefore, for any $|j_L\rangle$, we have$^1$
$$
P|j_L\rangle=\sum_i|i_L\rangle\langle i_L|j_L\rangle=|j_L\rangle.\tag2
$$
Now, if $\rho$ has eigendecomposition
$$
\rho=\sum_kp_k|\psi_k\rangle\langle\psi_k|\tag3
$$
then
$$
\sqrt{\rho}=\sum_k\sqrt{p_k}|\psi_k\rangle\langle\psi_k|.\tag4
$$
But $\rho$ is in the code subspace, so $|\psi_k\rangle=\sum_i\alpha_{ki}|i_L\rangle$ for some $\alpha_{ki}\in\mathbb{C}$. Substituting into $(4)$, we get
$$
\sqrt{\rho}=\sum_{ijk}\sqrt{p_k}\alpha_{ki}\overline{\alpha_{kj}}|i_L\rangle\langle j_L|.\tag5
$$
Finally, calculate
$$
\begin{align}
P\sqrt{\rho}&=P\sum_{ijk}\sqrt{p_k}\alpha_{ki}\overline{\alpha_{kj}}|i_L\rangle\langle j_L|\tag6\\
&=\sum_{ijk}\sqrt{p_k}\alpha_{ki}\overline{\alpha_{kj}}P|i_L\rangle\langle j_L|\tag7\\
&=\sum_{ijk}\sqrt{p_k}\alpha_{ki}\overline{\alpha_{kj}}|i_L\rangle\langle j_L|\tag8\\
&=\sqrt{\rho}\tag9
\end{align}
$$
where we used $(5)$, linearity of $P$, $(2)$, and finally $(5)$ again.
$^1$ Equivalently, one may take $(2)$ together with the requirement that $P$ vanishes on vectors orthogonal to the space spanned by $|i_L\rangle$ as the definition of the projector.